Distributed IC Power Delivery: Stability-Constrained Design Optimization and Workload-Aware Power Management

ABSTRACT Power delivery presents key design challenges in today’s systems ranging from high performance micro-processors to mobile systems-on-a-chips (SoCs). A robust power delivery system is essential to ensure reliable operation of on-die devices. Nowadays it has become an important design trend to place multiple voltage regulators on-chip in a distributive manner to cope with power supply noise. However, stability concern arises because of the complex interactions be-tween multiple voltage regulators and bulky network of the surrounding passive parasitics. The recently developed hybrid stability theorem (HST) is promising to deal with the stability of such system by efficiently capturing the effects of all interactions, however, large overdesign and hence severe performance degradation are caused by the intrinsic conservativeness of the underlying HST framework. To address such challenge, this dissertation first extends the HST by proposing a frequency-dependent system partitioning technique to substantially reduce the pessimism in stability evaluation. By systematically exploring the theoretical foundation of the HST framework, we recognize all the critical constraints under which the partitioning technique can be performed rigorously to remove conservativeness while maintaining key theoretical properties of the partitioned subsystems. Based on that, we develop an efficient stability-ensuring automatic design flow for large power delivery systems with distributed on-chip regulation. In use of the proposed approach, we further discover new design insights for circuit designers such as how regulator topology, on-chip decoupling capacitance, and the number of integrated voltage regulators can be optimized for improved system tradeoffs between stability and performances. Besides stability, power efficiency must be improved in every possible way while maintaining high power quality. It can be argued that the ultimate power integrity and efficiency may be best achieved via a heterogeneous chain of voltage processing starting from on-board switching voltage regulators (VRs), to on-chip switching VRs, and finally to networks of distributed on-chip linear VRs. As such, we propose a heterogeneous voltage regulation (HVR) architecture encompassing regulators with complimentary characteristics in response time, size, and efficiency. By exploring the rich heterogeneity and tunability in HVR, we develop systematic workload-aware control policies to adapt heterogeneous VRs with respect to workload change at multiple temporal scales to significantly improve system power efficiency while providing a guarantee for power integrity. The proposed techniques are further supported by hardware-accelerated machine learning prediction of non-uniform spatial workload distributions for more accurate HVR adaptation at fine time granularity. Our evaluations based on the PARSEC benchmark suite show that the proposed adaptive 3-stage HVR reduces the total system energy dissipation by up to 23.9% and 15.7% on average compared with the conventional static two-stage voltage regulation using off- and on-chip switching VRs. Compared with the 3-stage static HVR, our runtime control reduces system energy by up to 17.9% and 12.2% on average. Furthermore, the proposed machine learning prediction offers up to 4.1% reduction of system energy